International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)

Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com Volume 2, Issue 6, November December 2013 ISSN 2278-6856

Study on self organizing approach for moving object detection and tracking for visual surveillance
Ms. Neha Dhavale1 , Mrs. Seema Biday2
1,2

Department of Computer Engineering, Terna Engineering College, Nerul.

Abstract:

Visual surveillance is a very active research area in computer vision and moving object detection and tracking is often the first step in applications such as video surveillance. Here, we propose a methodology to detect and track human image based on self organization through artificial neural networks. We choose the HSV color space, relying on the hue, saturation and value properties of each color to represent each weight vector. A neural network mapping method is proposed to use a whole trajectory incrementally in time fed as an input to the network. The adopted artificial neural network is organized as a 2-D flat grid of neurons (or nodes). Each node computes a function of the weighted linear combination of incoming inputs, where weights resemble the neural network learning. Each node could be represented by a weight vector obtained ,collecting the weights related to incoming links. An incoming pattern is mapped to the node whose model is most similar (according to a predefined metric) to the pattern, and weight vectors in a neighborhood of such node are updated. We would focus on combining contour projection analysis with shape analysis to remove the shadow effect.

Keywords HSV, HMM, self organizing, moving objects.

1. INTRODUCTION
Visual surveillance systems include object detection, object classification, tracking, activity understanding, and semantic description. The scientific challenge is to devise and implement automatic systems able to detect and track moving objects, and interpret their activities and behaviors. behaviors. The detection of moving objects in video streams is the first relevant step of information extraction in many computer vision applications. Object tracking, in general, is a challenging problem. Difficulties in tracking objects can arise due to abrupt object motion, changing appearance patterns of the object and the scene, non rigid object structures, object-to-object and object-to-scene occlusions, and camera motion. Tracking is usually performed in the context of higher-level applications that require the location and/or shape of the object in every frame. Object tracking is an important task within the field field of computer vision. There are three key steps in video video analysis: detection of interesting moving objects, tracking of such objects from frame to frame, and analysis of object tracks to recognize their behavior. We studied a real-time visual tracking system on a controlled pan-tilt camera. The input/output HMM (Hidden Markov Model) Volume 2, Issue 6 November December 2013

is employed to model the overall visual tracking system in the spherical camera platform coordinate. In order to fast detect and track targets on a moving camera at the same time, the optical flow is adopted to observe the different displacement in the image sequence. A new efficient moving target detection method is proposed which is a improved background subtraction to detect moving objects. Two significant advantages were the improved the background subtraction and increased algorithm's running efficiency and offset sensitive deficiency of the light changes. Background subtraction is a widely used approach for detecting foreground objects in videos from a static camera. Indoor surveillance applications such as home-care and health-care monitoring, a motionless person should not be a part of the background. A reference background image without moving objects is, therefore, required for such applications. In this paper, an ICA (Independent Component Analysis)-based background subtraction scheme for foreground segmentation is presented. The proposed ICA model is based on the direct measurement of statistical independency that minimizes the difference between the joint PDF and the product of marginal PDFs, in which the probabilities are simply estimated from the relative frequency distributions. The proposed ICA model well performs the separation of highly-correlated signals. The moving object tracking method uses a one-dimensional optical flow under the rotating observer which avoids an ill-posed problem for calculating the optical flow based on the gradient method using one-dimensional brightness data. Convergence of SEOS (Simultaneous Estimation of Optical flow and State dynamics) was evaluated for both the Gauss-Seidel and Jacobi iterative techniques. It was shown that SEOS will converge for both Gauss-Seidel and Jacobi iterative schemes for any initial approximation. Also the effect of applying SEOS to a real-image sequence and tracking various objects within the Hamburg Taxi Sequence was examined. This object isolation feature of the SEOS technique could prove invaluable in numerous tracking applications. Background subtraction is an active researching field, because it can be used in many applications. An efficient background subtraction approach is the base which determines performance of the whole system. In this paper, the original GMM to improve its performance was simplified. Page 237

International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)

Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com Volume 2, Issue 6, November December 2013 ISSN 2278-6856
In the present work the concepts of dynamic template matching and frame differencing have been used to implement a robust automated single object tracking system. Using frame differencing on frame-by-frame basis a moving object, if any, is detected with high accuracy and efficiency. Once the object has been detected it is tracked by employing an efficient Template Matching algorithm. The templates used for the matching purposes are generated dynamically. This ensures that any change in the pose of the object does not hinder the tracking procedure. To automate the tracking process the camera is mounted on a pan-tilt arrangement, which is synchronized with a tracking algorithm. As and when the object being tracked moves out of the viewing range of the camera, the pan-tilt setup is automatically adjusted to move the camera so as to keep the object in view. Evaluation based on GT (Ground Truth) offers a framework for objective comparison of performance of alternate surveillance algorithms. Such evaluation techniques compare the output of the algorithm with the GT obtained manually by drawing bounding boxes around objects, or marking-up the pixel boundary of objects, or labeling objects of interest in the original video sequence. Manual generation of GT is an extraordinarily time-consuming and tedious task and, thus, inevitably error prone even for motivated researchers. Interpretation of evaluation results is obviously based on the type of GT used for comparison. However, established standards for GT are only just emerging. There are several ambiguities involved in the process of GT generation. For example, whether to account only for individual objects or also for groups of objects, or whether to look at the bounding boxes or exact shapes of objects. network. This makes the network structure much simpler and the learning process much more efficient. The neural network is organized as a 2-D flat grid of neurons (or nodes) and, similarly to self-organizing maps (SOMs) or Kohonen networks, allows to produce representations of training samples with lower dimensionality, at the same time preserving topological neighborhood relations of the input patterns (nearby outputs correspond to nearby input patterns). The proposed method can be explained diagrammatically as shown in figure 2.1

Figure 2.1: Flow chart for human motion detection The algorithm can be explained as follows:Input: pixel value pt in frame It ,t=0..Last frame Output: background/foreground binary mask value B(pt) 1. Initialize model C for pixel p0 and store it into A 2. for t=1, LastFrame 3. Find best match cm in C to current sample pt 4. if ( cm found) then 5. B(pt)=0 //background 6. update A in the neighborhood of cm 7. else if( pt shadow) then 8. B(pt)=0 //background 9. else 10. B(pt)=1 //foreground

2. PROPOSED METHOD
Visual surveillance systems include object detection, object classification, tracking, activity understanding, and semantic description. The detection of moving objects in video streams is the first relevant step of information extraction in many computer vision applications. Object tracking, in general, is a challenging problem. The proposed method is to adopt a biologically inspired problem-solving method based on visual attention mechanisms. The objective is to detect the objects that keep the user attention in accordance with a set of predefined features, including gray level, motion and shape shape features. The approach defines a method for the generation of an active attention focus to monitor dynamic dynamic scenes for surveillance purposes. The idea is to build the background model by learning in a selforganizing manner for many background variations, i.e., background motion cycles, seen as trajectories of pixels in time. Based on the background model through a map of motion and stationary patterns, the algorithm can detect motion and selectively update the background model. A neural network based method is proposed to use a whole trajectory incrementally in time fed as an input to the

3. Performance Evaluation
The proposed system can be experimented with different settings of adjustable parameters until the results seemed optimal over the entire sequence. Following parameters can be used for performance evaluation. a) Processing time We calculate the elapsed time using tic (Timer on) and toc (Timer off). tic and toc functions work together to measure elapsed time. We will evaluate the elapsed time for three methods viz. Frame subtraction, Background subtraction and the proposed method. Based on the processing time, we would determine which system is fastest. Page 238

Volume 2, Issue 6 November December 2013

International Journal of Emerging Trends & Technology in Computer Science (IJETTCS)

Web Site: www.ijettcs.org Email: editor@ijettcs.org, editorijettcs@gmail.com Volume 2, Issue 6, November December 2013 ISSN 2278-6856
b) Accuracy After classification of pixels as background & foreground pixels, we will retrieve color image from it. In retrieved image, we will check accuracy of retrieved object which is being tracked and represent it in percentage by comparing with original image. We can compare this accuracy measure with output of other object tracking algorithms. measuring accuracy there are different metrics, namely Recall, Precision, and Similarity. 1) Recall Recall also known as detection rate, gives the percentage of detected true positives as compared to the total number of true positives in the ground truth. Recall = (3.1) Where, tp = total number of true positives = total number of false negatives, and indicates the total number of items present in the ground truth. 2) Precision Precision, also known as positive prediction, gives the percentage of detected true positives as compared to the total number of items detected by the method. Precision = (3.2) Where, fp = total number of false positives = total number of detected items. Using the above mentioned metrics, generally, a method is considered good if it reaches high Recall values, without sacrificing Precision. 3) Similarity The pixel-based similarity measure is defined asSimilarity = (3.3) [2] R. T. Collins, A. J. Lipton, T. Kanade, H. Fujiyoshi, D. Duggins, Y. Tsin, D. Tolliver, N. Enomoto, O. Hasegawa, P. Burt, and L.Wixson, A system for video surveillance and monitoring, Tech.Rep. CMURI-TR-00-12, The Robotics Inst., Carnegie Mellon Univ., Pittsburgh, PA, 2000. [3] Alper Yilmaz, Omar Javed, Mubarak Shah, Object Tracking: A Survey, ACM Computing Surveys, Vol. 38, No. 4, Article 13, Publication date: December 2006. [4] Cheng-Ming Huang, Yi-Ru Chen , Li-Chen Fu,Real-Time Object Detection and Tracking on a Moving Camera Platform, ICCAS-SICE, National Conference,2009. [5] Lianqiang Niu, Nan Jiang ,A Moving Objects Detection Algorithm on Imporved Background Subtraction , Eighth International Confrence, Intelligent Systems Design and Applications,2008,ISDA 08 [6] Tsai DM, Lai SC, Independent Component Analysis-Based Background Subtraction for Indoor Surveillance, IEEE Trans Image Process, 2009 Jan; 18(1):158-67, doi: 10.1109/TIP.2008.2007558. [7] Kinoshit, K.Enokida, M.Izumid, M.Murakami, Tracking of a Moving Object Using OneDimensional Optical Flow with a Rotating Observer,9th International Conference ,Control, Automation, Robotics and Vision, 2006, ICARCV '06. [8] Bauer, N.J, Pathirana, P.N,Object Focused Simultaneous Estimation of Optical Flow and State Dynamics, International Conference, Intelligent Sensors, Sensor Networks and Information Processing, 2008,ISSNIP 2008. [9] Zhen Tang,Zhenjiang Miao ,Fast Background Subtraction and Shadow Elimination Using Improved Gaussian Mixture Model, Haptic, Audio and Visual Environments and Games, 2007. HAVE 2007, IEEE International Workshop. [10] Karan Gupta1, Anjali V. Kulkarni Implementation of an Automated Single Camera Object Tracking System Using Frame Differencing and Dynamic Template Matching, Indian Institute of Technology, Kanpur, India. [11] J. Black, T.J,Ellis, and P. Rosin, A NovelMethod for Video Tracking Performance valuation,IEEE Workshop on Performance Analysis of Video Surveillance and Tracking (PETS2003), pages 125 132, October 2003.

4. CONCLUSION
Currently, methods used in moving object detection are mainly the frame subtraction method, the background subtraction method and the optical flow method. The limitations of these methods are light changes, moving background, cast shadows, bootstrapping, and camouflage, poor anti-noise performance. A neural network based method is proposed to use a whole trajectory incrementally in time fed as an input to the network. The objective is to build the background model by learning in a self-organizing manner for many background variations, i.e., background motion cycles, seen as trajectories of pixels in time. We would focus on combining contour projection analysis with shape analysis to remove the shadow effect.

REFERENCES
[1] J. M. Ferryman, Ed, Evaluation of Tracking and Surveillance in Proc. 9th IEEE Int. Workshop on Performance, 2006.

Volume 2, Issue 6 November December 2013

Page 239